Odyssey is an AI lab building general-purpose world models, causal, multimodal systems that learn to predict and interact with the world over long horizons , rather than systems trained primarily on text. The relevant audience is robotics teams, game developers, simulation engineers, and researchers who need AI that reasons about physics, motion, and cause and effect from visual and sensor data. The launch matters because the company is announcing a $310 million Series B at a $1.45 billion valuation , led by Natural Capital, with Amazon, AMD Ventures , GV, In-Q-Tel and others participating, signaling that major strategic backers are now treating world models as a serious parallel track to LLMs.
The company was founded in 2023 by CEO Oliver Cameron and CTO Jeff Hawke, both of whom have deep experience in self-driving technology . Cameron previously co-founded autonomous vehicle startup Voyage, which was acquired by GM's Cruise , and Hawke helped build Wayve, where world models were used to train driving policies. That background shows up in the product line, which includes Odyssey-2, a general-purpose world model focused on physical accuracy, Starchild-1, a multimodal system that learns from richer interaction than vision alone, Agora-1, a multi-agent model that lets multiple humans or AIs share a single live simulation, and PROWL, a reinforcement learning framework where an agent explores environments to improve the underlying world model.
The pitch in the launch video is that current AI learns mostly from written records, while humans learn by moving through and bumping into the world, and that training models on that same kind of environmental data unlocks applications across robotics, gaming, science, education, and healthcare. With this round, Odyssey is positioning its models as infrastructure for simulating reality at scale, with the funding aimed at expanding training compute and pushing the Odyssey-2, Starchild, and Agora families toward real workloads in interactive video, game creation, and embodied AI.
Comments (12)
Sign in to join the discussion.
P
Priya RaghavanJun 18, 2026
World models is the kind of phrase that either ages like fine wine or like milk in a hot car. Rooting for the wine outcome.
T
Tomáš BřežnýJun 18, 2026
The tweet structure is clean but burying the investor list under a media link is criminal. That's the part the replies want to fight about.
L
Lina O.Jun 18, 2026
ok wait, raising at this size for world models before anyone agrees what a world model actually is is the most 2024 thing I've ever seen. Respect.
M
Margaux LefèvreJun 18, 2026
Curious what the compute burn looks like at this scale. World simulation isn't exactly known for its cozy unit economics.
F
Fadel MorenoJun 18, 2026
Skeptic hat on: every world model demo I've seen is a 6 second clip of grass swaying. Show me a cat knocking a glass off a counter and behaving like a cat.
K
kenjiJun 18, 2026
Building in adjacent sim space and genuinely happy to see this. Also slightly nauseous, which is how you know it's the right reaction.
Z
Zubair IqbalJun 18, 2026
Cool, but is there an API or do I have to email an account exec who replies in 9 business days? Asking for every developer ever.
S
Samira HaddadJun 18, 2026
Before my team can even evaluate: SOC2, data residency, and is there a way to run inference without sending frames to your cloud? Procurement needs to know.
N
Ngozi EzeJun 18, 2026
Language models predict the next token. World models predict the next moment. One of those sentences sells, the other ships.
B
Boris SalonenJun 18, 2026
Reminds me of the early bets we made in robotics sim. The thesis rhymes, the cap table looks heavy enough to mean it.
R
Reda K.Jun 18, 2026
Beyond language models is a bold flex when half the industry still can't get language models to spell strawberry. Love the confidence though.
V
Viktor CutsJun 18, 2026
The launch reel needed one more hard cut at the 0:08 mark, the b-roll lingers and you lose people right before the payoff. Otherwise the grade is gorgeous.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Odyssey
We're pioneering world models, to go beyond language models
We're pioneering world models, to go beyond language models
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Odyssey is an AI lab building general-purpose world models, causal, multimodal systems that learn to predict and interact with the world over long horizons , rather than systems trained primarily on text. The relevant audience is robotics teams, game developers, simulation engineers, and researchers who need AI that reasons about physics, motion, and cause and effect from visual and sensor data. The launch matters because the company is announcing a $310 million Series B at a $1.45 billion valuation , led by Natural Capital, with Amazon, AMD Ventures , GV, In-Q-Tel and others participating, signaling that major strategic backers are now treating world models as a serious parallel track to LLMs. The company was founded in 2023 by CEO Oliver Cameron and CTO Jeff Hawke, both of whom have deep experience in self-driving technology . Cameron previously co-founded autonomous vehicle startup Voyage, which was acquired by GM's Cruise , and Hawke helped build Wayve, where world models were used to train driving policies. That background shows up in the product line, which includes Odyssey-2, a general-purpose world model focused on physical accuracy, Starchild-1, a multimodal system that learns from richer interaction than vision alone, Agora-1, a multi-agent model that lets multiple humans or AIs share a single live simulation, and PROWL, a reinforcement learning framework where an agent explores environments to improve the underlying world model. The pitch in the launch video is that current AI learns mostly from written records, while humans learn by moving through and bumping into the world, and that training models on that same kind of environmental data unlocks applications across robotics, gaming, science, education, and healthcare. With this round, Odyssey is positioning its models as infrastructure for simulating reality at scale, with the funding aimed at expanding training compute and pushing the Odyssey-2, Starchild, and Agora families toward real workloads in interactive video, game creation, and embodied AI.
Comments (12)
World models is the kind of phrase that either ages like fine wine or like milk in a hot car. Rooting for the wine outcome.
The tweet structure is clean but burying the investor list under a media link is criminal. That's the part the replies want to fight about.
ok wait, raising at this size for world models before anyone agrees what a world model actually is is the most 2024 thing I've ever seen. Respect.
Curious what the compute burn looks like at this scale. World simulation isn't exactly known for its cozy unit economics.
Skeptic hat on: every world model demo I've seen is a 6 second clip of grass swaying. Show me a cat knocking a glass off a counter and behaving like a cat.
Building in adjacent sim space and genuinely happy to see this. Also slightly nauseous, which is how you know it's the right reaction.
Cool, but is there an API or do I have to email an account exec who replies in 9 business days? Asking for every developer ever.
Before my team can even evaluate: SOC2, data residency, and is there a way to run inference without sending frames to your cloud? Procurement needs to know.
Language models predict the next token. World models predict the next moment. One of those sentences sells, the other ships.
Reminds me of the early bets we made in robotics sim. The thesis rhymes, the cap table looks heavy enough to mean it.
Beyond language models is a bold flex when half the industry still can't get language models to spell strawberry. Love the confidence though.
The launch reel needed one more hard cut at the 0:08 mark, the b-roll lingers and you lose people right before the payoff. Otherwise the grade is gorgeous.